Efficient Recognition and Automatic Sorting Technology of Waste Textiles Based on Online Near infrared Spectroscopy and Convolutional Neural Network

• An intelligent, efficient, environmentally friendly and non-destructive identification and sorting technology for waste textiles is provided. • An online NIR qualitative identification model of 13 kinds of waste textiles is established by the convolutional neural network . • The accuracy of online identification and sorting for 13 kinds of waste textiles is above 95%. • The online recognition and sorting time of each sample is less than 2 s. In order to better recycle waste textiles and save resources, intelligent identification and sorting equipment and technology are urgently needed. In this work, an online near infrared (NIR) spectral library was established by utilizing self-developed online NIR device, including polyester, cotton, wool, silk, viscose, nylon, acrylic, polyester/cotton, polyester/wool, polyester/nylon, polyester/viscose, nylon/spandex and silk/cotton. Importantly, artificial intelligence technology was introduced into the identification and sorting of waste textiles, and two online NIR qualitative identification models covering above 13 kinds of waste textiles were constructed by the convolutional neural network (CNN) and Baidu deep learning platform PaddlePaddle. First, the input one-dimensional spectral data (901-2500 nm) was normalized and converted into a two-dimensional grayscale image of 40*40 pixels. Then feature extraction, compression and dimension reduction of multiple spectra were carried out through convolution and pooling. Finally, the category probability value of each kind of waste textiles was calculated by the CNN model and the maximum value was taken as the final classification of the fabric. Online identification tests were performed using 526 samples as an external validation set, presenting an accuracy of two CNN qualitative identification models were both more than 95.4%. In addition, the accuracy of online identification and sorting was above 95%, and the recognition and sorting time of each sample is less than 2 s, which can perform the efficient identification and automatic sorting of waste textiles.